We compare two diverse classification strategies on real-life biomedical data. One is based on a genetic algorithm-driven feature extraction method, combined with data fusion and the use of a simple, single classifier, such as linear discriminant analysis. The other exploits a single layer perceptron-based, data-driven evolution of the optimal classifier, and data fusion. We discuss the intricate interplay between dataset size, the number of features, and classifier complexity, and suggest different techniques to handle such problems.